Given that the prediction accuracy of the trans-ancestry PRS in the eMERGE African samples was relatively low compared with other populations, we performed further evaluation of the PRS-CSx-derived PRS in four UAB Black cohorts, namely REGARDS (1659 T2D cases, 5086 T2D controls), GenHAT (2776 cases, 2722 controls), HyperGEN (402 cases, 1494 controls) and WPC (300 cases, 355 controls) (Table 1). REGARDS, GenHAT and HyperGEN had largely consistent estimates of performance metrics, and the prediction accuracy was higher relative to the prediction in the eMERGE African samples. Specifically, the AUC of the PRS, adjusting for age, sex, and top 10 genetic PCs, was approximately 0.61, and the OR/SD estimates ranged from 1.70 to 1.85 across the three cohorts (Table 3; Additional File 2: Table S4). OR estimates contrasting individuals in the tail of the PRS distribution with the rest of the samples increased when more extreme cutoffs were applied; at the top 2% of the PRS distribution, an approximately three-fold increase in T2D risk was observed relative to the remaining 98% of the individuals (REGARDS: OR = 3.04, P value =
